Reading RDF in Apache Jena

This page details the setup of RDF I/O technology (RIOT) for Apache Jena.

See Writing RDF for details of the RIOT Writer system.

Full details of operations are given in the javadoc.


Much of the functionality is accessed via the Jena Model API; direct calling of the RIOT subsystem isn’t needed. A resource name with no URI scheme is assumed to be a local file name.

Applications typically use at most RDFDataMgr to read RDF datasets.

The major classes in the RIOT API are:

Class Comment
RDFDataMgr Main set of functions to read and load models and datasets
StreamRDF Interface for the output of all parsers
RDFParser Detailed setup of a parser
StreamManager Handles the opening of typed input streams
RDFLanguages Registered languages
RDFParserRegistry Registered parser factories

Determining the RDF syntax

The syntax of the RDF file is determined by the content type (if an HTTP request), then the file extension if there is no content type. Content type text/plain is ignored; it is assumed to be type returned for an unconfigured http server. The application can also pass in a declared language hint.

The string name traditionally used in is mapped to RIOT Lang as:

Jena reader RIOT Lang
"Turtle" TURTLE
"N3" N3

The following is a suggested Apache httpd .htaccess file:

AddType  text/turtle               .ttl
AddType  application/rdf+xml       .rdf
AddType  application/n-triples     .nt

AddType  application/ld+json       .jsonld

AddType  text/trig                 .trig
AddType  application/n-quads       .nq

AddType  application/trix+xml      .trix
AddType  application/rdf+thrift    .rt
AddType  application/rdf+protobuf  .rpb

Example 1 : Using the RDFDataMgr

RDFDataMgr provides operations to load, read and write models and datasets.

RDFDataMgr “load” operations create an in-memory container (model, or dataset as appropriate); “read” operations add data into an existing model or dataset.

// Create a model and read into it from file 
// "data.ttl" assumed to be Turtle.
Model model = RDFDataMgr.loadModel("data.ttl") ;

// Create a dataset and read into it from file 
// "data.trig" assumed to be TriG.
Dataset dataset = RDFDataMgr.loadDataset("data.trig") ;

// Read into an existing Model, "data2.ttl") ;

Example 2 : Common usage

The original Jena Model API operation for read and write provide another way to the same machinery:

Model model = ModelFactory.createDefaultModel() ;"data.ttl") ;

If the syntax is not as the file extension, a language can be declared:"", "TURTLE") ;

Example 3 : Using RDFParser

Detailed control over the setup of the parsing process is provided by RDFParser which provides a builder pattern. It has many options - see the javadoc for all details.

For example, to read Trig data, and set the error handler specially,

    Dataset dataset;
    // The parsers will do the necessary character set conversion.  
    try (InputStream in = new FileInputStream("data.some.unusual.extension")) {
        dataset = 


The parsers log to a logger called org.apache.jena.riot. To avoid WARN messages, set this to ERROR in the logging system of the application.

StreamManager and LocationMapper

Operations to read RDF data can be redirected to local copies and to other URLs. This is useful to provide local copies of remote resources.

By default, the RDFDataMgr uses the global StreamManager to open typed InputStreams. The StreamManager can be set using the RDFParser builder:

    // Create a copy of the global default StreamManager.
    StreamManager sm = StreamManager.get().clone();
    // Add directory "/tmp" as a place to look for files
    sm.addLocator(new LocatorFile("/tmp"));


It can also be set in a Context object given the the RDFParser for the operation, but normally this defaults to the global Context available via Context.get(). The constant SysRIOT.sysStreamManager, which is, is used.

Specialized StreamManagers can be configured with specific locators for data:

  • File locator (with own current directory)
  • URL locator
  • Class loader locator
  • Zip file locator

Configuring a StreamManager

The StreamManager can be reconfigured with different places to look for files. The default configuration used for the global StreamManager is a file access class, where the current directory is that of the java process, a URL accessor for reading from the web, and a class loader-based accessor. Different setups can be built and used either as the global set up, or on a per request basis.

There is also a LocationMapper for rewriting file names and URLs before use to allow placing known names in different places (e.g. having local copies of import http resources).

Configuring a LocationMapper

Location mapping files are RDF, usually written in Turtle although an RDF syntax can be used.

PREFIX lm: <>

[] lm:mapping
   [ lm:name "file:foo.ttl" ;      lm:altName "file:etc/foo.ttl" ] ,
   [ lm:prefix "file:etc/" ;       lm:altPrefix "file:ETC/" ] ,
   [ lm:name "file:etc/foo.ttl" ;  lm:altName "file:DIR/foo.ttl" ]

There are two types of location mapping: exact match renaming and prefix renaming. When trying to find an alternative location, a LocationMapper first tries for an exact match; if none is found, the LocationMapper will search for the longest matching prefix. If two are the same length, there is no guarantee on order tried; there is no implied order in a location mapper configuration file (it sets up two hash tables).

In the example above, file:etc/foo.ttl becomes file:DIR/foo.ttl because that is an exact match. The prefix match of file:/etc/ is ignored.

All string tests are done case sensitively because the primary use is for URLs.


  • Property values are not URIs, but strings. This is a system feature, not an RDF feature. Prefix mapping is name rewriting; alternate names are not treated as equivalent resources in the rest of Jena. While application writers are encouraged to use URIs to identify files, this is not always possible.
  • There is no check to see if the alternative system resource is equivalent to the original.

A LocationMapper finds its configuration file by looking for the following files, in order:

  • file:location-mapping.rdf
  • file:location-mapping.ttl
  • file:etc/location-mapping.rdf
  • file:etc/location-mapping.ttl

This is a specified as a path - note the path separator is always the character ‘;’ regardless of operating system because URLs contain ‘:’.

Applications can also set mappings programmatically. No configuration file is necessary.

The base URI for reading models will be the original URI, not the alternative location.

Advanced examples

Example code may be found in jena-examples:arq/examples.

Iterating over parser output

One of the capabilities of the RIOT API is the ability to treat parser output as an iterator, this is useful when you don’t want to go to the trouble of writing a full sink implementation and can easily express your logic in normal iterator style.

To do this you use AsyncParser.asyncParseTriples which parses the input on another thread:

    IteratorCloseable<Triple> iter = AsyncParser.asyncParseTriples(filename);
        // Do something with triple

Calling the iterator’s close method stops parsing and closes the involved resources. For N-Triples and N-Quads, you can use RiotParsers.createIteratorNTriples(input) which parses the input on the calling thread.

RIOT example 9.

Additional control over parsing is provided by the AsyncParser.of(...) methods which return AsyncParserBuilder instances. The builder features a fluent API that allows for fine-tuning internal buffer sizes as well as eventually obtaining a standard Java Stream. Calling the stream’s close method stops parsing and closes the involved resources. Therefore, these streams are best used in conjunction with try-with-resources blocks:

    try (Stream<Triple> stream = AsyncParser.of(filename)
            .setQueueSize(2).setChunkSize(100).streamTriples().limit(1000)) {
        // Do something with the stream

The AsyncParser also supports parsing RDF into a stream of EltStreamRDF elements. Each element can hold a triple, quad, prefix, base IRI or exception. For all Stream-based methods there also exist Iterator-based versions:

    IteratorCloseable<EltStreamRDF> it = AsyncParser.of(filename).asyncParseElements();
    try {
        while (it.hasNext()) {
            EltStreamRDF elt =;
            if (elt.isTriple()) {
               // Do something with elt.getTriple();
            } else if (elt.isPrefix()) {
               // Do something with elt.getPrefix() and elt.getIri();
    } finally {

Filter the output of parsing

When working with very large files, it can be useful to process the stream of triples or quads produced by the parser so as to work in a streaming fashion.

See RIOT example 4

Add a new language

The set of languages is not fixed. A new language, together with a parser, can be added to RIOT as shown in RIOT example 6